[
https://issues.apache.org/jira/browse/HDFS-2288?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13103296#comment-13103296
]
Konstantin Shvachko commented on HDFS-2288:
-------------------------------------------
Todd, visible length is a notion which exists when pipeline is in progress. As
Nicholas said it guarantees consistency of reads from replicas at different
stages.
In your scenario if I understood it correctly after writing one replica and
failing on hflush(), nobody would've seen any bytes yet, therefore visible
length should be 0.
When DN restarts after the crash there is no visible length since there is no
pipeline. Non-finalized replicas become RWR, which means the recovery hasn't
started for them yet. And clients should not read from RWR replicas until the
recovery finishes.
Section 8.3 of the design doc says "The length of an rwr is a negative number."
Also worth checking the replica state transition diagram under 9.1
So I think the behavior is correct (to the specs). What is the HBase logic
there?
> Replicas awaiting recovery should return a full visible length
> --------------------------------------------------------------
>
> Key: HDFS-2288
> URL: https://issues.apache.org/jira/browse/HDFS-2288
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: data-node
> Affects Versions: 0.23.0
> Reporter: Todd Lipcon
> Assignee: Todd Lipcon
> Priority: Critical
> Fix For: 0.23.0
>
> Attachments: hdfs-2288.txt
>
>
> Currently, if the client calls getReplicaVisibleLength for a RWR, it returns
> a visible length of 0. This causes one of HBase's tests to fail, and I
> believe it's incorrect behavior.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
